LiYuan/dysts

---
license: cc-by-4.0
task_categories:
- time-series-forecasting
tags:
- time
- multivariate
- forecasting
- univariate-time-series-forecasting
- multivariate-time-series-forecasting
pretty_name: Chaos Multivariate Time Series
size_categories:
- 1M<n<10M
---

### Chaotic Time Series Dataset

Multivariate time series from chaotic dynamical systems.

+ Each multivariate time series is a drawn from one chaotic dynamical system over an extended duration, making this dataset suitable for long-horizon forecasting tasks.

+ There are 4 million total multivariate observations, grouped into 135 systems and three granularities

+ The subdirectories `coarse`, `medium`, and `fine` each contain 135 `.csv` files, each of which contains a single multivariate time series of length 10,000

+ The number of channels varies depending on the specific dynamical system.

+ The time series are stationary due to the ergodic property of chaotic systems.

## Reference

For more information, or if using this code for published work, please cite the accompanying papers.

> William Gilpin. "Chaos as an interpretable benchmark for forecasting and data-driven modelling" Advances in Neural Information Processing Systems (NeurIPS) 2021 https://arxiv.org/abs/2110.05266

> William Gilpin. "Model scale versus domain knowledge in statistical forecasting of chaotic systems" Physical Review Research 2023 https://arxiv.org/abs/2303.08011

## Code

For executable code, or to simulate new trajectories, please see the [dysts repository on GitHub](https://github.com/williamgilpin/dysts)